# -*- coding: utf-8 -*- """laya-opencv安装脚本 / laya-opencv installer. 步骤：虚拟环境 → PyTorch → Laya → 检查 → 视觉组件（OpenCV，可选）→ 收尾。 由 install.bat / install.sh 调用，Windows、Linux 通用。可以重复运行，已装好的部分会跳过。 Called by install.bat / install.sh. Safe to re-run: finished steps are skipped. python app/install.py install python app/install.py --dry-run only print the commands """ import json import os import shutil import subprocess import sys from pathlib import Path sys.path.insert(0, str(Path(__file__).resolve().parent)) import wb_lang # noqa: E402 ROOT = Path(__file__).resolve().parent.parent VENV = ROOT / ".venv" IS_WINDOWS = os.name == "nt" VENV_PY = VENV / ("Scripts/python.exe" if IS_WINDOWS else "bin/python") DRY_RUN = "--dry-run" in sys.argv DEFAULTS = { "language": "auto", "pip_index": "auto", "torch_index": "", "install_local_laya": True, "install_vision": True, "desktop_shortcut": True, } # 视觉组件：OpenCV 5 的主包去掉了训练模块（cv2.ml），所以固定用 4.x VISION_PACKAGES = ["opencv-python-headless>=4.9,<5", "numpy"] # 每条消息: (中文, English) MSG = { "cfg_bad": ("[!] config.json 格式有误，已改用默认配置: %s", "[!] config.json is not valid JSON; using defaults: %s"), "cmd_failed": ("\n[!] 上面的命令失败了（退出码 %d），安装中止。", "\n[!] The command above failed (exit code %d); installation stopped."), "torch_have": (" 已安装 torch %s（GPU %s），跳过。", " torch %s is already installed (GPU %s); skipping."), "gpu_yes": ("可用", "available"), "gpu_no": ("不可用", "not available"), "torch_index": (" 按 config.json 的 torch_index 安装 PyTorch。", " Installing PyTorch from torch_index in config.json."), "torch_gpu": (" 检测到 NVIDIA 显卡，安装 GPU 版 PyTorch（体积约 2~3 GB，需要一些时间）。", " NVIDIA GPU found; installing the GPU build of PyTorch (about 2-3 GB, this takes a while)."), "torch_cpu": (" 没有检测到 NVIDIA 显卡，安装 CPU 版 PyTorch。", " No NVIDIA GPU found; installing the CPU build of PyTorch."), "torch_fallback": (" [!] 自动选择 PyTorch 版本失败，改用 pip 默认版本安装。", " [!] Could not pick a PyTorch build automatically; installing pip's default build."), "shortcut_name": ("laya-opencv", "laya-opencv"), "shortcut_done": (" 已在桌面创建快捷方式「%s」。", ' Created the desktop shortcut "%s".'), "title": (" laya-opencv 安装程序", " laya-opencv installer"), "dir": (" 安装目录: %s", " Install folder: %s"), "need_py": ("[!] 需要 Python 3.10 或更高版本，当前是 %s。", "[!] Python 3.10 or newer is required; this is %s."), "s_venv": ("准备虚拟环境 (.venv)", "Prepare the virtual environment (.venv)"), "venv_have": (" 已存在，直接使用。", " Already there; using it."), "venv_fail": ("[!] 创建虚拟环境失败。", "[!] Could not create the virtual environment."), "venv_apt": ("\n Debian / Ubuntu 请先执行: sudo apt install python3-venv", "\n On Debian / Ubuntu run first: sudo apt install python3-venv"), "s_torch": ("安装 PyTorch", "Install PyTorch"), "s_laya": ("安装 Laya 和服务端组件", "Install Laya and the server components"), "s_check": ("检查安装结果", "Check the installation"), "dry_skip": (" (dry-run，跳过检查)", " (dry run, check skipped)"), "check_fail": ("[!] 检查没有通过：laya=%s，服务端组件=%s，torch=%s", "[!] Check failed: laya=%s, server components=%s, torch=%s"), "gpu_ok": (" GPU 可用 (%s)", " GPU available (%s)"), "gpu_wrong_build": (" GPU 不可用 —— 有 NVIDIA 显卡，但当前 PyTorch 不是 GPU 版。\n" " Laya 会用 CPU 运行（能用，只是慢一些）。想用 GPU，\n" " 请到 https://pytorch.org/get-started/locally/ 选择对应命令，\n" " 用 %s -m pip 重新安装 torch。", " GPU not available: there is an NVIDIA GPU, but this PyTorch is not a GPU build.\n" " Laya will run on the CPU (it works, just slower). To use the GPU,\n" " pick the matching command at https://pytorch.org/get-started/locally/\n" " and reinstall torch with %s -m pip."), "gpu_none": (" GPU 不可用，使用 CPU 运行。", " GPU not available; running on the CPU."), "remote_only": ("\n config.json 里 install_local_laya 为 false：跳过 PyTorch 和 Laya，\n" " 工作台只使用远程接口（TypeSafe / aiask.me）。", "\n install_local_laya is false in config.json: skipping PyTorch and Laya.\n" " The workbench will use remote interfaces only (TypeSafe / aiask.me)."), "s_vision": ("安装视觉组件（OpenCV）", "Install the vision components (OpenCV)"), "vision_have": (" 已安装 OpenCV %s，跳过。", " OpenCV %s is already installed; skipping."), "vision_swap": (" 已安装的 OpenCV %s 不带训练模块，换成 4.x 版本。", " The installed OpenCV %s has no training module; switching to a 4.x build."), "vision_ok": (" OpenCV %s，训练模块可用。", " OpenCV %s, training module available."), "vision_fail": (" [!] 视觉组件没有装上，图像检测和训练暂时不可用，其余功能不受影响。\n" " 解决网络问题后重新运行安装脚本即可补装。", " [!] The vision components could not be installed; image inspection and training are unavailable for now.\n" " Everything else works. Run the installer again once the network problem is fixed."), "vision_skip": ("\n config.json 里 install_vision 为 false：跳过 OpenCV，不提供图像检测和训练。", "\n install_vision is false in config.json: skipping OpenCV; image inspection and training will be unavailable."), "s_finish": ("收尾", "Finish up"), "done": (" 完成。", " Done."), "all_done": (" 安装完成。运行 %s 启动工作台。", " Installation finished. Run %s to start the workbench."), "first_start": (" 首次启动会下载模型（约 650 MB），以后启动只需加载模型。", " The first start downloads the model (about 650 MB); later starts only load it."), } def load_config(): cfg = dict(DEFAULTS) path = ROOT / "config.json" example = ROOT / "config.example.json" if not path.exists() and example.exists() and not DRY_RUN: path.write_bytes(example.read_bytes()) # 首次安装：从模板生成 config.json if not path.exists() and example.exists(): path = example if path.exists(): try: cfg.update(json.loads(path.read_text(encoding="utf-8-sig"))) except ValueError as error: print("[!] config.json: %s" % error) return cfg CFG = load_config() LANG = wb_lang.detect(CFG.get("language")) # "auto"：中文环境用清华镜像，其他环境用官方源 if str(CFG.get("pip_index") or "").strip().lower() == "auto": CFG["pip_index"] = "https://pypi.tuna.tsinghua.edu.cn/simple" if LANG == "zh" else "" def T(key, *args): text = MSG[key][1 if LANG == "en" else 0] return text % args if args else text def step(n, total, key): print("\n[%d/%d] %s" % (n, total, T(key))) def run(cmd, check=True): """执行命令并把输出直接显示出来，返回是否成功。""" cmd = [str(c) for c in cmd] print(" > " + " ".join(cmd)) if DRY_RUN: return True code = subprocess.call(cmd, cwd=str(ROOT)) if code != 0 and check: raise SystemExit(T("cmd_failed", code)) return code == 0 def probe(code): """在虚拟环境里跑一小段 Python，返回输出；失败返回 None。""" if DRY_RUN or not VENV_PY.exists(): return None try: out = subprocess.run([str(VENV_PY), "-c", code], capture_output=True, text=True, timeout=180, cwd=str(ROOT)) except (OSError, subprocess.TimeoutExpired): return None return out.stdout.strip() if out.returncode == 0 else None def pip(*args, check=True): cmd = [VENV_PY, "-m", "pip", "install", "--disable-pip-version-check"] if CFG.get("pip_index"): cmd += ["-i", CFG["pip_index"]] return run(cmd + list(args), check=check) def has_nvidia(): exe = shutil.which("nvidia-smi") if not exe: return False try: return subprocess.run([exe], capture_output=True, timeout=30).returncode == 0 except (OSError, subprocess.TimeoutExpired): return False def torch_state(): """返回 (版本, cuda 是否可用)；没装返回 (None, False)。""" out = probe("import torch; print(torch.__version__); print(int(torch.cuda.is_available()))") if not out: return None, False lines = out.splitlines() return lines[0], lines[-1] == "1" def install_torch(nvidia): version, cuda = torch_state() if version: print(T("torch_have", version, T("gpu_yes") if cuda else T("gpu_no"))) return if CFG.get("torch_index"): print(T("torch_index")) run([VENV_PY, "-m", "pip", "install", "--disable-pip-version-check", "torch", "--index-url", CFG["torch_index"]]) return # uv 能按显卡驱动自动挑对应的 PyTorch 版本；没有显卡时装体积小得多的 CPU 版 backend = "auto" if nvidia else "cpu" print(T("torch_gpu") if nvidia else T("torch_cpu")) if pip("uv", check=False): cmd = [VENV_PY, "-m", "uv", "pip", "install", "--python", VENV_PY, "--torch-backend", backend] if CFG.get("pip_index"): cmd += ["--default-index", CFG["pip_index"]] if run(cmd + ["torch"], check=False): return print(T("torch_fallback")) pip("torch") def vision_state(): """返回 (OpenCV 版本, 训练模块是否可用)；没装返回 (None, False)。""" out = probe("import cv2, numpy; print(cv2.__version__); " "print(int(all(hasattr(getattr(cv2, 'ml', None), n) for n in ('SVM_create', 'RTrees_create', 'KNearest_create'))))") if not out: return None, False lines = out.splitlines() return lines[0], lines[-1] == "1" def install_vision(): version, has_ml = vision_state() if version and has_ml: print(T("vision_have", version)) return if version: print(T("vision_swap", version)) run([VENV_PY, "-m", "pip", "uninstall", "-y", "opencv-python", "opencv-python-headless"], check=False) ok = pip(*VISION_PACKAGES, check=False) version, has_ml = vision_state() if DRY_RUN: return if ok and version and has_ml: print(T("vision_ok", version)) else: print(T("vision_fail")) def make_shortcut(): if not IS_WINDOWS or DRY_RUN or not CFG.get("desktop_shortcut", True): return name = T("shortcut_name") target = str(ROOT / "start.bat").replace("'", "''") workdir = str(ROOT).replace("'", "''") script = ( "$d=[Environment]::GetFolderPath('Desktop');" "$s=(New-Object -ComObject WScript.Shell).CreateShortcut((Join-Path $d '%s.lnk'));" "$s.TargetPath='%s';$s.WorkingDirectory='%s';$s.Save()" % (name, target, workdir) ) # 项目原来叫 Laya 工作台 / Laya Workbench。桌面上的旧快捷方式如果指向这个目录， # 或者指向的位置已经不存在（文件夹改过名），就删掉，免得留下一个失效的 cleanup = ( "$d=[Environment]::GetFolderPath('Desktop');$w=New-Object -ComObject WScript.Shell;" "foreach($n in @('Laya \u5de5\u4f5c\u53f0','Laya Workbench')){$f=Join-Path $d ($n+'.lnk');" "if(Test-Path -LiteralPath $f){$t=$w.CreateShortcut($f).TargetPath;" "if(($t -eq '%s') -or (($t) -and -not (Test-Path -LiteralPath $t))){Remove-Item -LiteralPath $f}}}" % target ) try: done = subprocess.run(["powershell.exe", "-NoProfile", "-ExecutionPolicy", "Bypass", "-Command", script], capture_output=True, timeout=60) if done.returncode == 0: print(T("shortcut_done", name)) subprocess.run(["powershell.exe", "-NoProfile", "-ExecutionPolicy", "Bypass", "-Command", cleanup], capture_output=True, timeout=60) except (OSError, subprocess.TimeoutExpired): pass def main(): local = bool(CFG.get("install_local_laya", True)) vision = bool(CFG.get("install_vision", True)) total = 2 + (3 if local else 0) + (1 if vision else 0) print("=" * 56) print(T("title")) print(T("dir", ROOT)) print("=" * 56) if sys.version_info < (3, 10): raise SystemExit(T("need_py", sys.version.split()[0])) step(1, total, "s_venv") if VENV_PY.exists(): print(T("venv_have")) else: if not run([sys.executable, "-m", "venv", VENV], check=False): raise SystemExit(T("venv_fail") + ("" if IS_WINDOWS else T("venv_apt"))) run([VENV_PY, "-m", "ensurepip", "--upgrade"], check=False) if local: step(2, total, "s_torch") nvidia = has_nvidia() install_torch(nvidia) step(3, total, "s_laya") pip("-U", "laya[serve]") step(4, total, "s_check") if DRY_RUN: print(T("dry_skip")) else: laya_version = probe("import laya; print(laya.__version__)") serve_ok = probe("import laya.serve, fastapi, uvicorn; print('ok')") version, cuda = torch_state() if not laya_version or not serve_ok or not version: raise SystemExit(T("check_fail", laya_version, serve_ok, version)) print(" laya %s" % laya_version) print(" torch %s" % version) if cuda: gpu = probe("import torch; print(torch.cuda.get_device_name(0))") print(T("gpu_ok", gpu or "cuda")) elif nvidia: print(T("gpu_wrong_build", ".venv\\Scripts\\python.exe" if IS_WINDOWS else ".venv/bin/python")) else: print(T("gpu_none")) else: print(T("remote_only")) if vision: step(total - 1, total, "s_vision") install_vision() else: print(T("vision_skip")) step(total, total, "s_finish") make_shortcut() print(T("done")) print("\n" + "=" * 56) print(T("all_done", "start.bat" if IS_WINDOWS else "./start.sh")) if local: print(T("first_start")) print("=" * 56) if __name__ == "__main__": main()